THE 20 BEST CHATGPT PROMPTS
You do not need 500 prompts. You need to understand the few that compound — and the structure underneath them. Here are twenty fully-built prompts for the work that actually matters, and the anatomy to write your own.
By Editorial · Published Jun 25, 2026 · 20 min read
On This Page
The internet is drowning in lists of ChatGPT prompts — five hundred here, a thousand there, every one a single vague line like "write a sales page" or "create a business plan." They are nearly useless, and for a specific reason: an unconstrained one-liner gives the model no role, no context, and no standard to hit, so it returns the bland average of everything it has read. The number of prompts in a list is a vanity metric. What actually changes your output is understanding the small number of prompts that compound and the structure underneath all of them. This guide is twenty professional prompts for the work founders and operators actually do — thinking, deciding, writing, analyzing, building — each written out in full, plus the anatomy that lets you write your own for anything else.
This is a working resource, not a swipe file to hoard. Every prompt below is complete and ready to paste; the only thing you add is your own specifics in the bracketed slots.
The anatomy of a prompt that works
Almost every strong prompt, regardless of task, shares the same five parts, and learning them is worth more than any list. A good prompt opens with a role that tells the model which specialist to become, gives it the context of the situation and goal, states the task precisely, names the deliverables it must return in the format you want, and imposes constraints that block weak or dishonest output. The official guidance from the model makers says the same thing in plainer terms: be specific, give context, and show the model the format you expect, as OpenAI lays out in its prompt engineering guide. The constraints are the part most people omit and the part that matters most — "take a position," "do not invent statistics," "flag what you are unsure of" are what separate a usable answer from a confident, generic one.
The prompts are reusable because they run on a small set of variables. Replace these tokens with your own specifics before running any prompt.
| Variable | Replace with | Example |
|---|---|---|
[TASK] | What you need done | Cut our onboarding from 5 steps to 3 |
[CONTEXT] | The situation and goal | Seed-stage SaaS, churn is too high |
[AUDIENCE] | Who the output is for | Our engineering lead |
[GOAL] | The outcome that matters | A decision we can ship this week |
[INPUT] | Material to work from | Pasted notes, code, or a draft |
These tokens are intentional fill-ins, not unfinished sections. The twenty prompts are grouped into five kinds of work — thinking, communicating, analyzing, building, and growing — and the most valuable group is the first, because the highest-leverage use of a large language model is not writing your emails, it is pressure-testing your reasoning.
Think before you act
The prompts most people never use are the ones that pay off most: using the model as a reasoning partner that decomposes problems and attacks your own conclusions. None of these write anything you will publish; all of them make your decisions better.
1. First-principles problem decomposition
This prompt breaks a tangled problem down to its fundamentals instead of reasoning by analogy from how others have solved it. It forces the model to separate the facts from the assumptions, which is usually where a stuck problem is actually stuck.
You are a strategist who reasons from first principles, not by analogy.
CONTEXT
- The problem: [TASK].
- Background: [CONTEXT].
TASK
Decompose this problem to its fundamentals.
DELIVERABLES
1. The problem restated as plainly as possible.
2. The hard facts that are genuinely true regardless of how things are "usually done".
3. The assumptions hiding inside my framing of the problem - and which are worth challenging.
4. The problem rebuilt from the facts upward, ignoring convention.
5. Two or three approaches that fall out of that rebuild, ranked by leverage.
CONSTRAINTS
- Separate fact from assumption explicitly; do not let a convention pass as a fact.
- Challenge my framing where it is weak rather than solving the problem as I posed it.
- Prefer the non-obvious approach the first-principles view exposes.
2. Decision under uncertainty
This prompt turns a fuzzy "what should I do" into a structured decision: real options, the tradeoffs of each, a recommendation, and — most usefully — what evidence would change the answer. The last part is what keeps it honest rather than confidently wrong.
You are a decision advisor helping me choose under uncertainty.
CONTEXT
- The decision: [TASK].
- What is at stake and the constraints: [CONTEXT].
- The outcome I care about most: [GOAL].
TASK
Structure this decision and recommend a path.
DELIVERABLES
1. The real options, including any I may have missed and the "do nothing" option.
2. For each: the main upside, the main risk, and what it costs to reverse if wrong.
3. A clear recommendation and the reasoning behind it.
4. The single piece of evidence that would most change this recommendation.
5. The cheapest test I could run before committing.
CONSTRAINTS
- Distinguish what is known from what is being assumed.
- Weight reversibility heavily; favor reversible bets when uncertainty is high.
- Give a real recommendation, not a list of considerations.
3. Red-team and pre-mortem
This prompt attacks a plan before reality does. It imagines the plan has already failed and works backwards to the most likely causes, then names the cheapest check that would expose the biggest risk now. Used on your own thesis, it is the closest thing to a free insurance policy.
You are a red-team analyst running a pre-mortem on my plan.
CONTEXT
- The plan or decision: [TASK].
- What I am counting on for it to work: [CONTEXT].
TASK
Assume it is [TIMEFRAME] later and this has clearly failed. Work backwards.
DELIVERABLES
1. The three most likely reasons it failed, ranked by damage.
2. The early warning sign for each - what I would see before it is too late.
3. The assumption whose failure would be most catastrophic.
4. The single cheapest action now that most reduces the biggest risk.
CONSTRAINTS
- Attack the plan's logic and evidence, not a strawman of it.
- Be concrete about failure modes; "execution risk" is not an answer.
- If the plan is genuinely sound, say which risks are acceptable rather than inventing problems.
4. Prioritization under load
This prompt sorts a overwhelming task list by what actually matters rather than what feels urgent. It applies a real framework — importance against urgency — and forces a decision on what to drop, which is the part most people avoid.
You are a chief of staff helping me prioritize a heavy workload.
CONTEXT
- My tasks: [INPUT].
- What success this week looks like: [GOAL].
TASK
Prioritize these using an importance-versus-urgency framework.
DELIVERABLES
1. Each task sorted into: do now (important and urgent), schedule (important, not urgent), delegate (urgent, not important), or drop (neither).
2. The one task that, if done, makes several others easier or unnecessary.
3. What I should explicitly NOT do this week, and why that is safe.
4. A realistic order for the "do now" items.
CONSTRAINTS
- Be willing to put things in "drop"; a prioritization with nothing dropped is useless.
- Tie importance to the stated goal, not to how loud a task feels.
- Flag anything that looks like busywork disguised as progress.
Write and communicate
These handle the writing that operators actually do day to day — not marketing copy, which has its own content prompt library, but the high-stakes message, the brief, and the explanation.
5. The difficult message
This prompt drafts the email or message you have been avoiding — pushing back, delivering bad news, declining, negotiating. It gives you strategically different versions so you can choose the stance, rather than one take you have to accept or rewrite.
You are a communications advisor helping me write a high-stakes message.
CONTEXT
- The situation: [CONTEXT].
- Who I am writing to and our relationship: [AUDIENCE].
- What I need to achieve: [GOAL].
TASK
Draft this message in two or three strategically different versions.
DELIVERABLES
For each version: a label for its stance (e.g. firm, collaborative, conciliatory), the full message, and one line on when to choose it.
CONSTRAINTS
- Be direct and respectful; no corporate hedging or passive-aggression.
- Preserve the relationship without surrendering the point.
- Keep each version tight - say the hard thing clearly and stop.
6. Executive brief from raw notes
This prompt compresses messy notes into a decision-ready brief that leads with the answer. It is built for a reader with two minutes, which forces the clarity that long reports hide.
You are a chief of staff turning raw notes into a decision brief for [AUDIENCE].
CONTEXT
- The decision or update: [CONTEXT].
- My raw material: [INPUT].
TASK
Write a tight brief, not a report.
DELIVERABLES
1. Bottom line up front: the recommendation or headline in two sentences.
2. The three points that most support it, one line each.
3. The strongest counter-point, stated fairly.
4. The decision being asked for and the next concrete step.
CONSTRAINTS
- Lead with the answer; never make the reader hunt for it.
- Cut anything that does not change the decision.
- Mark any number that is an estimate rather than a known figure.
- Keep it under 250 words.
7. Article with a point of view
This prompt drafts a piece that argues something rather than surveying a topic neutrally. For deeper, channel-specific writing systems this is only a starting point — the full treatment lives in the content creation prompt library — but for a fast, opinionated draft it does the job.
You are a writer drafting an article with a real point of view.
CONTEXT
- Topic: [TASK].
- Audience: [AUDIENCE].
- My angle: [ANGLE, or "propose the most defensible one"].
TASK
Write a complete first draft.
DELIVERABLES
1. A headline that promises a specific payoff, not a vague topic.
2. An opening that states the actual argument in the first few sentences.
3. Body sections, each making one point backed by a concrete example or detail.
4. A conclusion that lands the idea and says what to do or think next.
CONSTRAINTS
- Take a position; a piece any competitor could have written is a failure.
- No filler phrases ("in today's fast-paced world", "game-changing").
- Do not invent statistics, studies, or quotes; flag where a real source is needed.
8. Plain-language explainer
This prompt makes the model teach a complex thing simply, the Feynman way — which also exposes where its own understanding (or yours) is thin. It is the fastest way to learn something well enough to make a decision about it.
You are a brilliant teacher who explains hard things simply.
CONTEXT
- Concept to explain: [TASK].
- My current level: [CONTEXT, e.g. "smart but new to this field"].
TASK
Explain this so I genuinely understand it.
DELIVERABLES
1. The core idea in two plain sentences, no jargon.
2. A concrete analogy or example that makes it click.
3. The one thing most people get wrong about it.
4. A check: a question I should be able to answer if I understood.
CONSTRAINTS
- No jargon without immediately defining it in plain words.
- Prefer one good example over three shallow ones.
- If the concept has a genuinely hard part, do not paper over it - name it.
Analyze and research
These extract signal from documents and markets. For deep, source-disciplined research workflows there is a dedicated research prompt library; the four here cover the everyday analytical tasks — and they carry the same rule: the model proposes, you verify.
9. Document synthesizer
This prompt turns a long document into a decision-useful summary that stays faithful to the source. Its key constraint is that it works only from what you paste and flags anything it cannot find, rather than filling gaps from memory.
You are an analyst synthesizing a document for a busy decision-maker.
CONTEXT
- The document: [INPUT].
- What I need to decide or understand: [GOAL].
TASK
Synthesize it into a decision-useful summary.
DELIVERABLES
1. The core message in three sentences.
2. The points that matter for my goal, in order of importance.
3. Anything surprising, contradictory, or weakly supported in the document.
4. What it does NOT address that I would need to know.
CONSTRAINTS
- Summarize only what is in the document; if something is not there, say so rather than inferring it.
- Quote sparingly and only to preserve a precise meaning.
- Separate the document's claims from your own interpretation.
10. Rigorous SWOT
This prompt does a SWOT analysis that is actually useful, because it forces evidence behind each entry and bans the generic filler ("strength: good team") that makes most SWOTs worthless.
You are a strategy consultant running a rigorous SWOT analysis.
CONTEXT
- The company or initiative: [TASK].
- What we are deciding: [GOAL].
TASK
Produce a SWOT where every entry earns its place.
DELIVERABLES
Strengths, weaknesses, opportunities, and threats - for each item, the specific evidence or reasoning behind it, and why it matters for the decision.
CONSTRAINTS
- No generic entries; "strong team" or "competition" without specifics is banned.
- Distinguish what is verifiable from what is assumed; label assumptions.
- End with the single most important implication for the decision, not just the grid.
11. Competitor snapshot
This prompt profiles a competitive field fast, while refusing to invent the precise figures models love to hallucinate. It separates what is publicly known from what it is inferring, and tells you how to check the rest.
You are a competitive analyst building a fast, honest snapshot of [INDUSTRY / CATEGORY].
CONTEXT
- Our vantage point: [CONTEXT].
- What we need to decide: [GOAL].
TASK
Map the competitive field.
DELIVERABLES
1. The most relevant players, grouped by type (incumbent, challenger, niche).
2. For each: their wedge, who they serve, and their most visible weakness.
3. The underserved gap no one owns well.
4. The competitor most likely to be underestimated, and why.
CONSTRAINTS
- Do not invent funding, revenue, or customer numbers; write "unverified" and note how to check.
- Separate what is publicly verifiable from what you are inferring.
- Tie the analysis to our decision, not to a generic feature comparison.
12. Meeting notes to decisions
This prompt turns a wall of meeting notes into the only things that matter afterward: decisions made, actions owned, and questions left open. It is the difference between a meeting that happened and a meeting that produced anything.
You are a chief of staff converting meeting notes into an action record.
CONTEXT
- The notes: [INPUT].
TASK
Extract what actually came out of this meeting.
DELIVERABLES
1. Decisions made, stated unambiguously.
2. Action items, each with an owner and a due date if one was given (mark "unassigned" if not).
3. Open questions that were raised but not resolved.
4. Anything that was discussed at length but produced no decision - flagged for follow-up.
CONSTRAINTS
- Do not invent owners, dates, or decisions that were not in the notes.
- Mark anything ambiguous as needing confirmation rather than guessing.
- Keep it to what is actionable; cut the chatter.
Build
These are the technical prompts, written to hold the model to real engineering standards rather than letting it emit plausible-looking code with no specification behind it.
13. Code from a spec
This prompt produces code against an explicit specification, which is the difference between getting what you need and getting something that compiles. It forces the requirements to be stated before a line is written.
You are a senior engineer writing production-quality code.
CONTEXT
- What it must do: [TASK].
- Language and environment: [CONTEXT].
- Constraints (performance, dependencies, style): [GOAL, or "sensible defaults"].
TASK
Write the code, then explain the key decisions.
DELIVERABLES
1. The complete, runnable code.
2. Inline comments only where a decision is non-obvious.
3. Edge cases handled, and any you deliberately did not - stated explicitly.
4. How to test it.
CONSTRAINTS
- Restate the requirements in one line before coding, so a wrong assumption surfaces early.
- Handle errors and edge cases; do not write happy-path-only code.
- If a requirement is ambiguous, state the assumption you made rather than guessing silently.
14. Debugger
This prompt makes the model diagnose before it patches — finding the root cause rather than suppressing the symptom. The constraint to explain the cause is what stops it from handing you a fix you do not understand.
You are a senior engineer debugging a problem methodically.
CONTEXT
- The code: [INPUT].
- What it does versus what it should do: [CONTEXT].
- Error message, if any: [GOAL].
TASK
Find the root cause and fix it.
DELIVERABLES
1. The actual root cause, explained - not just the symptom.
2. The corrected code.
3. Why the original failed, so I understand it.
4. Anything nearby that is likely to break for the same reason.
CONSTRAINTS
- Diagnose before fixing; do not suppress the symptom and call it solved.
- If you cannot be certain of the cause from what I gave you, say what additional information would confirm it.
- Do not silently rewrite unrelated parts of the code.
15. Code review
This prompt reviews code the way a careful senior would — security, correctness, and edge cases first, style last — and prioritizes findings so you fix what matters before what is merely tidy.
You are a senior engineer reviewing a change before it merges.
CONTEXT
- The code or diff: [INPUT].
- What it is supposed to do: [CONTEXT].
TASK
Review it and return prioritized findings.
DELIVERABLES
For each issue: its severity (blocker / important / minor), what is wrong, and the concrete fix. Order by severity.
CONSTRAINTS
- Lead with correctness, security, and edge cases; style comes last.
- Distinguish a real bug from a preference, and label which is which.
- If the change is solid, say so rather than manufacturing nitpicks.
16. Technical documentation
This prompt writes documentation a real user could follow, organized around what they are trying to do rather than around the code's internal structure. It bans the worst documentation sin: describing what the code is instead of how to use it.
You are a technical writer documenting code for the people who will use it.
CONTEXT
- The code: [INPUT].
- Who reads this: [AUDIENCE].
TASK
Write usable documentation.
DELIVERABLES
1. A one-paragraph overview: what it does and when to reach for it.
2. How to use it, organized around tasks the reader wants to accomplish.
3. Inputs, outputs, and key options, in plain terms.
4. The common mistakes or gotchas and how to avoid them.
CONSTRAINTS
- Organize around what the reader is trying to do, not around the code's structure.
- Show realistic usage, not toy examples.
- Do not document behavior you cannot see in the provided code.
Grow
The last group is for the career and learning work the audience does for themselves and their teams — kept honest, because the temptation to fabricate is highest here.
17. Skill-learning roadmap
This prompt builds a realistic path to a new skill, sequenced from fundamentals to application, with a way to check progress. It is structured to prevent the usual failure of learning plans, which is collapsing into a list of resources with no order.
You are an expert tutor designing a learning path for a motivated adult.
CONTEXT
- Skill I want: [TASK].
- My starting point: [CONTEXT].
- Time I can give it: [GOAL].
TASK
Design a realistic roadmap.
DELIVERABLES
1. The sequence of concepts from foundation to application, in order.
2. For each stage: what to learn, and a small project that proves I learned it.
3. The fastest path to being useful, even before mastery.
4. The mistakes beginners make in this skill, and how to skip them.
CONSTRAINTS
- Sequence it; do not hand me an undifferentiated list of resources.
- Favor doing over consuming - every stage should have an output.
- Be honest about what genuinely takes time and cannot be shortcut.
18. Interview preparation
This prompt prepares you for a specific role by generating the questions you will actually face and pressuring your answers, rather than handing you generic platitudes. It includes the hard questions, not just the easy ones.
You are an interview coach preparing me for a specific role.
CONTEXT
- The role: [TASK].
- My background: [CONTEXT].
TASK
Prepare me properly.
DELIVERABLES
1. The 8-10 questions most likely for this specific role, including the hard ones.
2. For two or three of the hardest, a strong answer structure I could adapt.
3. The weakness in my background most likely to be probed, and how to address it honestly.
4. Three sharp questions I should ask them.
CONSTRAINTS
- Tailor to this role, not to generic interview advice.
- Do not script answers I would have to fake; build structures around my real experience.
- Be honest about a real weakness rather than pretending I have none.
19. Experience translator
This prompt turns your real experience into resume or profile language that lands — without inventing anything. The hard constraint is that it works only from what you actually did, which is both ethical and what survives a reference check.
You are a resume writer translating real experience into strong, honest language.
CONTEXT
- The role I am targeting: [TASK].
- What I actually did: [INPUT].
TASK
Rewrite my experience to land with a hiring manager.
DELIVERABLES
1. Bullet points that lead with outcome and impact, grounded in what I actually did.
2. Where a real metric exists in what I gave you, use it; where it does not, write impact qualitatively.
3. The single strongest line, positioned first.
CONSTRAINTS
- Invent nothing - no fabricated metrics, titles, or achievements.
- If a claim would need a number I did not provide, do not make one up; phrase it honestly.
- Cut vague verbs ("responsible for", "helped with") in favor of what I specifically did.
20. Weekly operating plan
This prompt turns goals into an actual week, allocating time to what matters and protecting it from the urgent-but-trivial. It closes the loop between intention and the calendar, where most plans quietly die.
You are a productivity coach turning my goals into a realistic week.
CONTEXT
- My goals for this period: [INPUT].
- My fixed commitments and constraints: [CONTEXT].
TASK
Build a weekly plan that actually moves the goals.
DELIVERABLES
1. The two or three outcomes that, if achieved this week, make it a success.
2. Time blocked for the important work, protected from the urgent-but-trivial.
3. What to say no to this week to make room.
4. A simple end-of-week check to see if it worked.
CONSTRAINTS
- Be realistic about capacity; an over-packed plan is a failed plan.
- Protect deep-work time explicitly rather than hoping it appears.
- Tie every block to one of the stated goals.
Why the "master prompt" does not work
Most mega-lists end with a magic incantation: "act as a team of world-class experts, ask clarifying questions, optimize for accuracy, efficiency, and actionable results." Skip it. Asking a model to be every expert at once and to optimize for everything gives it no specific role and no real constraints, so it produces a confident, shapeless answer that is good at nothing in particular. Vagueness scales badly: the broader the instruction, the more the model falls back on the generic average. A specific role with hard constraints — the structure in every prompt above — beats a grand incantation every time, which is the whole reason these twenty are built the way they are. This is the practical core of working with generative AI: you get back exactly as much precision as you put in.
For the deeper, domain-specific systems, this guide is the front door to three specialist libraries: the same structural approach applied to web design, to technology research, and to content creation. The pattern is identical; only the role and constraints change.
The Bottom Line
A list of five hundred prompts is a list of five hundred ways to get a mediocre answer. What actually compounds is the anatomy — role, context, task, deliverables, constraints — and the discipline to use the model for thinking, not just typing. The twenty prompts here cover the work that matters, but the real takeaway is the structure underneath them: once you can see it, you can write a strong prompt for anything in seconds, and you will never need a list of five hundred again. The model is the fast, capable, occasionally unreliable collaborator. You are the one who sets the standard it has to meet.
What makes a ChatGPT prompt good?+
A good prompt assigns the model a specific role, gives it real context, names the exact deliverables, and imposes constraints that block generic or fabricated output. Structure matters far more than length or clever wording — the same five-part anatomy works across nearly every task.
Do I really need hundreds of prompt templates?+
No. A small set of well-built prompts that you adapt with variables will outperform a giant library of vague one-liners. Understanding the underlying structure lets you write a strong prompt for any task in seconds, which is more useful than memorizing five hundred of them.
What is the variable framework in these prompts?+
It is a set of bracketed tokens like [TASK], [CONTEXT], [GOAL], and [AUDIENCE] that you replace with your own specifics before running a prompt. It turns one prompt into a reusable template that works across any project.
Does ChatGPT make up facts and sources?+
Yes. Language models generate plausible text from patterns rather than retrieving verified records, so they can fabricate citations, statistics, and quotes that look real. Any factual output should be verified against a primary source before you rely on it, which is why these prompts are built to flag uncertainty rather than hide it.
Are these prompts only for ChatGPT?+
No. The same structure works on any current frontier model. The prompts are written to be model-agnostic because what makes them effective is the role, context, and constraints, not anything specific to one product.